Re: [PATCH v2 1/2] hw/nvme: add support for boot partiotions

[Keith Busch]

On Tue, Jun 01, 2021 at 09:07:56PM +0200, Klaus Jensen wrote:
> On Jun  1 11:07, Keith Busch wrote:
> > On Tue, Jun 01, 2021 at 07:41:34PM +0200, Klaus Jensen wrote:
> > > On Jun  1 10:19, Keith Busch wrote:
> > > > On Tue, Jun 01, 2021 at 08:07:48PM +0530, Gollu Appalanaidu wrote:
> > > > > NVMe Boot Partitions provides an area that may be read by the host
> > > > > without initializing queues or even enabling the controller. This
> > > > > allows various platform initialization code to be stored on the NVMe
> > > > > device instead of some separete medium.
> > > > >
> > > > > This patch adds the read support for such an area, as well as support
> > > > > for updating the boot partition contents from the host through the
> > > > > FW Download and Commit commands.
> > > >
> > > > Please provide some details on what platform initilization sequence
> > > > running on QEMU is going to make use of this feature.
> > > >
> > > 
> > > I totally get your reluctance to accept useless features like device
> > > self-test and ill-supported ones like write uncorrectable.
> > > 
> > > But I think this feature qualifies just fine for the device. It is useful
> > > for embedded development and while there might not be any qemu boards that
> > > wants to use this *right now*, it allows for experimentation. And this is 
> > > a
> > > feature that actually *is* implemented by real products for embedded
> > > systems.
> > 
> > That wasn't my request, though. I am well aware of the feature and also
> > have hardware that implements it. It just sounds like you haven't
> > actually tested this feature under the protocol's intended use cases
> > inside this environment. I think that type of testing and a high level
> > description of it in the changelog ought to be part of acceptance
> > criteria.
> > 
> 
> Alright, I see.
> 
> You'd like to see this tested by defining a new board that loads firmware
> over PCIe from the device?

Yes, something like that.

When the feature was initially published, I took a very brief look at
how qemu could use it and concluded this wasn't very practical here. I
would be happy to know if there's any example platform that can use it,
though. That, to me, demostrates sufficient value.